In [1]:

import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.linear_model import LinearRegression
import pandas as pd

import warnings
warnings.filterwarnings("ignore")

import numpy as np import matplotlib.pyplot as plt import seaborn as sns from sklearn.linear_model import LinearRegression import pandas as pd import warnings warnings.filterwarnings("ignore")

In [2]:

%run ./module_P2_utils.ipynb

%run ./module_P2_utils.ipynb

In [3]:

# df = pd.read_csv("./tmp/df_03_quali.csv")
df = pd.read_pickle("./tmp/df_03_quali.gzip")

# df = pd.read_csv("./tmp/df_03_quali.csv") df = pd.read_pickle("./tmp/df_03_quali.gzip")

Analysis focused on Quantitative features¶

In [4]:

sns.displot(data=df, hue='nutriscore_grade', x='nutriscore_score', height=10)

sns.displot(data=df, hue='nutriscore_grade', x='nutriscore_score', height=10)

Out[4]:

<seaborn.axisgrid.FacetGrid at 0x7f6518319100>

In [5]:

sns.displot(data=df, hue='nutriscore_grade', x='nutriscore_score',kind='ecdf', height=10)

sns.displot(data=df, hue='nutriscore_grade', x='nutriscore_score',kind='ecdf', height=10)

Out[5]:

<seaborn.axisgrid.FacetGrid at 0x7f657b7f9a00>

We can see nutriscore grade and score may be correlated, but some values are overlapping, so we may expect it difficult to make any projections thanks to it.

Correlations between features¶

Let's use the Pearson Correlation coefficient for our analysis of correlation

For convenience and further analysis, we use absolute values.

In [6]:

correlations = df.select_dtypes('float').corr(method="pearson")
correlations = np.abs(correlations)
correlations

correlations = df.select_dtypes('float').corr(method="pearson") correlations = np.abs(correlations) correlations

Out[6]:

	additives_n	nutriscore_score	energy-kcal_100g	energy_100g	fat_100g	saturated-fat_100g	carbohydrates_100g	sugars_100g	proteins_100g	salt_100g	sodium_100g	nutrition-score-fr_100g
additives_n	1.000000	0.190463	0.071431	0.093514	0.072095	0.136920	0.178192	0.161574	0.019512	0.111632	0.111617	0.295182
nutriscore_score	0.190463	1.000000	0.514698	0.578317	0.665420	0.613793	0.242434	0.182480	0.039898	0.284028	0.284035	1.000000
energy-kcal_100g	0.071431	0.514698	1.000000	0.967918	0.704853	0.561154	0.530614	0.229329	0.397283	0.266112	0.266121	0.366877
energy_100g	0.093514	0.578317	0.967918	1.000000	0.731808	0.575507	0.567980	0.249249	0.426725	0.287250	0.287248	0.401920
fat_100g	0.072095	0.665420	0.704853	0.731808	1.000000	0.751979	0.197373	0.158031	0.466066	0.410888	0.410883	0.448844
saturated-fat_100g	0.136920	0.613793	0.561154	0.575507	0.751979	1.000000	0.148300	0.245602	0.396218	0.350177	0.350170	0.429570
carbohydrates_100g	0.178192	0.242434	0.530614	0.567980	0.197373	0.148300	1.000000	0.433564	0.095892	0.046727	0.046714	0.206927
sugars_100g	0.161574	0.182480	0.229329	0.249249	0.158031	0.245602	0.433564	1.000000	0.000709	0.008917	0.008917	0.191569
proteins_100g	0.019512	0.039898	0.397283	0.426725	0.466066	0.396218	0.095892	0.000709	1.000000	0.490700	0.490705	0.147951
salt_100g	0.111632	0.284028	0.266112	0.287250	0.410888	0.350177	0.046727	0.008917	0.490700	1.000000	0.999988	0.280558
sodium_100g	0.111617	0.284035	0.266121	0.287248	0.410883	0.350170	0.046714	0.008917	0.490705	0.999988	1.000000	0.280559
nutrition-score-fr_100g	0.295182	1.000000	0.366877	0.401920	0.448844	0.429570	0.206927	0.191569	0.147951	0.280558	0.280559	1.000000

In [7]:

mask = np.zeros_like(correlations)
mask[np.triu_indices_from(mask)] = True
sns.axes_style("white")
plt.subplots(figsize=(20,20))
sns.heatmap(correlations, mask=mask, vmax=1, square=True, xticklabels=correlations.columns, yticklabels=correlations.columns, annot=True, cmap='cividis')

mask = np.zeros_like(correlations) mask[np.triu_indices_from(mask)] = True sns.axes_style("white") plt.subplots(figsize=(20,20)) sns.heatmap(correlations, mask=mask, vmax=1, square=True, xticklabels=correlations.columns, yticklabels=correlations.columns, annot=True, cmap='cividis')

Out[7]:

<AxesSubplot:>

We consider a good correlation if corr > 0.5, and a pretty good correlation if corr > 0.3.

We can see that most correlated values are

• energy and energy-kcal
• nutriscore_score and nutriscore-score-fr
• sodium and salt

which was indeed predictable and doesn't teach us something

But, interestingly, we can see that nutriscore is well correlated with:

• saturated_fat
• fat
• energy
• energy_kcal
• nova_group

Also we can see some of the nutriments are correlated each others. By extension we may find out that nutriscore is a combination of all those nutriments.

In [8]:

plt.figure(figsize=(20, 50))
sns.pairplot(df, y_vars=['nutriscore_score'], x_vars=df.select_dtypes('float') )

plt.figure(figsize=(20, 50)) sns.pairplot(df, y_vars=['nutriscore_score'], x_vars=df.select_dtypes('float') )

Out[8]:

<seaborn.axisgrid.PairGrid at 0x7f6477e0b250>

<Figure size 1440x3600 with 0 Axes>

---

---

In [9]:

X = df.dropna().loc[:,'fat_100g'].values.reshape(-1, 1)
y = df.dropna().loc[:,'nutriscore_score'].values.reshape(-1, 1)

X = df.dropna().loc[:,'fat_100g'].values.reshape(-1, 1) y = df.dropna().loc[:,'nutriscore_score'].values.reshape(-1, 1)

In [10]:

reg = LinearRegression().fit(X, y)

reg = LinearRegression().fit(X, y)

In [11]:

reg.coef_

reg.coef_

Out[11]:

array([[0.47787512]])

Score is the $ R² $ defined as $ (1 - \frac{u}{v})$, where $ u $ is the residual sum of squares ((y_true - y_pred)** 2).sum() and $v$ is the total sum of squares ((y_true - y_true.mean()) ** 2).sum()

formula

$ R² = 1 - \frac{\sum{yi - f(yi)}} {\sum{yi - \bar{y}}}$

In [12]:

reg.score(X, y)

reg.score(X, y)

Out[12]:

0.3501325468540978

In that case, R² is not very good

In [13]:

nutriscore_pred = reg.predict(X)
plt.scatter(df.dropna()['fat_100g'], nutriscore_pred, color='red')

nutriscore_pred = reg.predict(X) plt.scatter(df.dropna()['fat_100g'], nutriscore_pred, color='red')

Out[13]:

<matplotlib.collections.PathCollection at 0x7f644e29c1f0>

We can try to make a linear regression with multiple features

In [14]:

X = df.dropna()[['fat_100g', 'energy_100g', 'sugars_100g', 'salt_100g']].to_numpy()
y = df.dropna()['nutriscore_score'].to_numpy().reshape(-1, 1)

X = df.dropna()[['fat_100g', 'energy_100g', 'sugars_100g', 'salt_100g']].to_numpy() y = df.dropna()['nutriscore_score'].to_numpy().reshape(-1, 1)

In [15]:

reg = LinearRegression().fit(X, y)

reg = LinearRegression().fit(X, y)

In [16]:

reg.coef_

reg.coef_

Out[16]:

array([[ 3.95715190e-01, -1.66460882e-03,  4.24424202e-01,
         4.25859762e+00]])

In [17]:

reg.score(X, y)

reg.score(X, y)

Out[17]:

0.5630929468277883

Here, the R² score is better, we are above 0.5

In [17]:

---

---

MISC Analyses¶

In [18]:

pd.crosstab(df['nutriscore_grade'], df['pnns_groups_1'] )

pd.crosstab(df['nutriscore_grade'], df['pnns_groups_1'] )

Out[18]:

pnns_groups_1	Alcoholic beverages	Beverages	Cereals and potatoes	Composite foods	Fat and sauces	Fish Meat Eggs	Fruits and vegetables	Milk and dairy products	Salty snacks	Sugary snacks	unknown
nutriscore_grade
a	1	4472	31515	6948	2695	12563	31083	9058	1750	1216	13582
b	1	8780	14229	13802	4166	12321	5237	14663	2427	2595	15439
c	10	9710	15029	16982	21122	17372	7256	16224	8723	12859	29400
d	21	7383	11844	12244	18644	29238	1046	39221	12268	49794	41966
e	11	15097	2211	1574	9704	16569	114	5011	5301	57864	18470

In [19]:

plt.figure(figsize=(15,15))
plt.title("mean nutriscore for pnns_group and nutriscore_grade")
sns.heatmap(df[['pnns_groups_1','nutriscore_grade', 'nutriscore_score']].pivot_table(index='pnns_groups_1', columns='nutriscore_grade', values='nutriscore_score', aggfunc=np.mean), annot=True)

plt.figure(figsize=(15,15)) plt.title("mean nutriscore for pnns_group and nutriscore_grade") sns.heatmap(df[['pnns_groups_1','nutriscore_grade', 'nutriscore_score']].pivot_table(index='pnns_groups_1', columns='nutriscore_grade', values='nutriscore_score', aggfunc=np.mean), annot=True)

Out[19]:

<AxesSubplot:title={'center':'mean nutriscore for pnns_group and nutriscore_grade'}, xlabel='nutriscore_grade', ylabel='pnns_groups_1'>

In [20]:

df.info()

df.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 1986648 entries, 0 to 1988475
Data columns (total 39 columns):
 #   Column                                   Dtype              
---  ------                                   -----              
 0   code                                     object             
 1   creator                                  object             
 2   created_datetime                         datetime64[ns, UTC]
 3   last_modified_datetime                   datetime64[ns, UTC]
 4   product_name                             object             
 5   brands                                   object             
 6   brands_tags                              object             
 7   categories                               object             
 8   categories_tags                          object             
 9   categories_en                            object             
 10  countries                                object             
 11  countries_tags                           object             
 12  countries_en                             object             
 13  ingredients_text                         object             
 14  additives_n                              float64            
 15  ingredients_from_palm_oil_n              category           
 16  ingredients_that_may_be_from_palm_oil_n  category           
 17  nutriscore_score                         float64            
 18  nutriscore_grade                         object             
 19  nova_group                               category           
 20  pnns_groups_1                            object             
 21  pnns_groups_2                            object             
 22  states                                   object             
 23  states_tags                              object             
 24  states_en                                object             
 25  main_category                            object             
 26  main_category_en                         object             
 27  energy-kcal_100g                         float64            
 28  energy_100g                              float64            
 29  fat_100g                                 float64            
 30  saturated-fat_100g                       float64            
 31  carbohydrates_100g                       float64            
 32  sugars_100g                              float64            
 33  proteins_100g                            float64            
 34  salt_100g                                float64            
 35  sodium_100g                              float64            
 36  nutrition-score-fr_100g                  float64            
 37  countries_cleaned                        object             
 38  countries_cleaned_cat                    category           
dtypes: category(4), datetime64[ns, UTC](2), float64(12), object(21)
memory usage: 619.6+ MB

---

---

Analysis focused on Qualitative features pnns¶

In [21]:

fig, axes = plt.subplots(1, 3, figsize=(20, 15))

for i, col in enumerate(df[['nutriscore_grade', 'pnns_groups_1', 'pnns_groups_2']]):
    ax = axes[i % 3]
    vals = df[col].value_counts(normalize=True)
    ax.pie(vals, labels=vals.index, startangle=30)
    ax.set_title(col)

fig.subplots_adjust(wspace=.2)

fig, axes = plt.subplots(1, 3, figsize=(20, 15)) for i, col in enumerate(df[['nutriscore_grade', 'pnns_groups_1', 'pnns_groups_2']]): ax = axes[i % 3] vals = df[col].value_counts(normalize=True) ax.pie(vals, labels=vals.index, startangle=30) ax.set_title(col) fig.subplots_adjust(wspace=.2)

Here we can see that pnns_group1 is mainly filled with 'unknown' values (about 60%)

In [22]:

sns.displot(df, y='pnns_groups_1', x='nutriscore_grade', height=10, cmap='viridis', hue='nutriscore_grade')

sns.displot(df, y='pnns_groups_1', x='nutriscore_grade', height=10, cmap='viridis', hue='nutriscore_grade')

Out[22]:

<seaborn.axisgrid.FacetGrid at 0x7f644c04c130>

We can see that for having a healthy meals (cat a), one would prefer to eat fruits, vegetables, cereals and potatoes. At the opposite, having sugary snack is the worste (cat e) meal.

In [23]:

plt.figure(figsize=(20, 20))
sns.jointplot(data=df, x='nutriscore_score', y='pnns_groups_1')

plt.figure(figsize=(20, 20)) sns.jointplot(data=df, x='nutriscore_score', y='pnns_groups_1')

Out[23]:

<seaborn.axisgrid.JointGrid at 0x7f644beb0040>

<Figure size 1440x1440 with 0 Axes>

In [24]:

pd.crosstab(df['nutriscore_grade'], df['pnns_groups_2'] )

pd.crosstab(df['nutriscore_grade'], df['pnns_groups_2'] )

Out[24]:

pnns_groups_2	Alcoholic beverages	Appetizers	Artificially sweetened beverages	Biscuits and cakes	Bread	Breakfast cereals	Cereals	Cheese	Chocolate products	Dairy desserts	...	Salty and fatty products	Sandwiches	Soups	Sweetened beverages	Sweets	Teas and herbal teas and coffees	Unsweetened beverages	Vegetables	Waters and flavored waters	unknown
nutriscore_grade
a	1	205	71	553	4433	2287	18745	608	9	1140	...	205	308	396	70	641	2	0	21021	3296	13582
b	1	573	1525	939	5681	1270	5312	917	21	1031	...	615	714	1311	979	1583	214	1533	2094	717	15439
c	10	3076	1109	5336	6217	2818	4563	3990	142	3027	...	2754	1119	823	897	7098	202	1474	1883	119	29400
d	21	5576	1212	22170	3796	2283	5011	32385	2065	824	...	5500	1648	40	2250	23109	468	1261	291	50	41966
e	11	1393	1057	27968	293	165	1486	3721	11155	102	...	3675	131	11	8846	16249	251	3278	33	36	18470

5 rows × 40 columns

In [25]:

sns.displot(df, y='pnns_groups_2', x='nutriscore_grade', height=10, cmap='viridis', hue='nutriscore_grade')

sns.displot(df, y='pnns_groups_2', x='nutriscore_grade', height=10, cmap='viridis', hue='nutriscore_grade')

Out[25]:

<seaborn.axisgrid.FacetGrid at 0x7f644bc469d0>

Correlations

In [26]:

pd.crosstab(df['nutriscore_grade'], df['pnns_groups_1'])

pd.crosstab(df['nutriscore_grade'], df['pnns_groups_1'])

Out[26]:

pnns_groups_1	Alcoholic beverages	Beverages	Cereals and potatoes	Composite foods	Fat and sauces	Fish Meat Eggs	Fruits and vegetables	Milk and dairy products	Salty snacks	Sugary snacks	unknown
nutriscore_grade
a	1	4472	31515	6948	2695	12563	31083	9058	1750	1216	13582
b	1	8780	14229	13802	4166	12321	5237	14663	2427	2595	15439
c	10	9710	15029	16982	21122	17372	7256	16224	8723	12859	29400
d	21	7383	11844	12244	18644	29238	1046	39221	12268	49794	41966
e	11	15097	2211	1574	9704	16569	114	5011	5301	57864	18470

---

---

Analysis on Qualitative feature Countries¶

In [27]:

for i, cat in enumerate(['FR', 'US', 'ES', 'IT']):
    subset = df[df.countries_cleaned_cat == cat][df.nutriscore_score != 0].dropna(subset=['nutriscore_score'])
    getStatisticsNutriscore(subset, cat)

for i, cat in enumerate(['FR', 'US', 'ES', 'IT']): subset = df[df.countries_cleaned_cat == cat][df.nutriscore_score != 0].dropna(subset=['nutriscore_score']) getStatisticsNutriscore(subset, cat)

--------------------
FR
(277289, 39)
std         8.692927e+00
skew        1.199459e-02
kurtosis   -8.777998e-01
mean        9.924822e+00
median      1.100000e+01
var         7.556697e+01
mad         7.405893e+00
prod                 inf
sum         2.752044e+06
Name: nutriscore_score, dtype: float64

--------------------
US
(226081, 39)
std         9.072824e+00
skew       -1.082389e-01
kurtosis   -9.258780e-01
mean        9.495663e+00
median      1.100000e+01
var         8.231613e+01
mad         7.724602e+00
prod                -inf
sum         2.146789e+06
Name: nutriscore_score, dtype: float64

--------------------
ES
(52610, 39)
std         8.387235e+00
skew        1.725937e-01
kurtosis   -8.296991e-01
mean        9.174358e+00
median      9.000000e+00
var         7.034571e+01
mad         7.230973e+00
prod                -inf
sum         4.826630e+05
Name: nutriscore_score, dtype: float64

--------------------
IT
(16988, 39)
std         8.460771e+00
skew       -2.604370e-02
kurtosis   -8.645842e-01
mean        9.380504e+00
median      1.100000e+01
var         7.158465e+01
mad         7.227253e+00
prod                 inf
sum         1.593560e+05
Name: nutriscore_score, dtype: float64

In [28]:

plt.figure(figsize=(15,15))
tmpdf = df.countries_cleaned.isin(['FR', 'US', 'ES', 'IT'])
sns.histplot(df[tmpdf], x='countries_cleaned', hue='nutriscore_grade')

plt.figure(figsize=(15,15)) tmpdf = df.countries_cleaned.isin(['FR', 'US', 'ES', 'IT']) sns.histplot(df[tmpdf], x='countries_cleaned', hue='nutriscore_grade')

Out[28]:

<AxesSubplot:xlabel='countries_cleaned', ylabel='Count'>

---

---

Categorization of Nutriscore_grade and correlations¶

In [29]:

df['nutriscore_grade_cat'] = df['nutriscore_grade'].astype('category')
df['nutriscore_grade_cat'].dropna().cat.categories
df['nutriscore_grade_cat_codes'] = df['nutriscore_grade_cat'].dropna().cat.codes
df['nutriscore_grade_cat'].dropna().cat.codes.unique()

df['nutriscore_grade_cat'] = df['nutriscore_grade'].astype('category') df['nutriscore_grade_cat'].dropna().cat.categories df['nutriscore_grade_cat_codes'] = df['nutriscore_grade_cat'].dropna().cat.codes df['nutriscore_grade_cat'].dropna().cat.codes.unique()

Out[29]:

array([3, 1, 0, 2, 4], dtype=int8)

In [30]:

pd.crosstab(df['countries_cleaned_cat'], df['nutriscore_grade'])

pd.crosstab(df['countries_cleaned_cat'], df['nutriscore_grade'])

Out[30]:

nutriscore_grade	a	b	c	d	e
countries_cleaned_cat
::	0	1	0	0	0
AD	10	20	18	24	8
AE	16	19	13	16	17
AF	16	22	35	41	24
AH	0	0	1	1	1
...	...	...	...	...	...
美国	0	0	1	0	0
西班	0	0	0	1	0
香港	0	4	1	4	0
대한	1	2	0	4	3
프랑	1	0	1	1	0

314 rows × 5 columns

In [31]:

df[['nutriscore_grade_cat_codes', 'nutriscore_score']].corr()

df[['nutriscore_grade_cat_codes', 'nutriscore_score']].corr()

Out[31]:

	nutriscore_grade_cat_codes	nutriscore_score
nutriscore_grade_cat_codes	1.000000	0.944612
nutriscore_score	0.944612	1.000000

In [31]:

---

---

Correlations and graphics¶

In [32]:

plt.figure(figsize=(15,15))
plt.scatter(x=df['energy_100g'], y=df['energy-kcal_100g'], alpha=0.8)

plt.figure(figsize=(15,15)) plt.scatter(x=df['energy_100g'], y=df['energy-kcal_100g'], alpha=0.8)

Out[32]:

<matplotlib.collections.PathCollection at 0x7f644b8fb8b0>

In [33]:

plt.figure(figsize=(15,15))
plt.scatter(x=df['nutriscore_grade'].astype('category').cat.codes, y=df['nutriscore_score'])

plt.figure(figsize=(15,15)) plt.scatter(x=df['nutriscore_grade'].astype('category').cat.codes, y=df['nutriscore_score'])

Out[33]:

<matplotlib.collections.PathCollection at 0x7f64b35b5f40>

In [34]:

plt.figure(figsize=(15,15))
plt.scatter(x=df['fat_100g'], y=df['nutriscore_score'], c=df['nutriscore_grade'].astype('category').cat.codes)

plt.figure(figsize=(15,15)) plt.scatter(x=df['fat_100g'], y=df['nutriscore_score'], c=df['nutriscore_grade'].astype('category').cat.codes)

Out[34]:

<matplotlib.collections.PathCollection at 0x7f64b35238e0>

In [35]:

plt.figure(figsize=(15,15))
plt.scatter(x=df['fat_100g'], y=df['energy_100g'], c=df['nutriscore_grade'].astype('category').cat.codes)

plt.figure(figsize=(15,15)) plt.scatter(x=df['fat_100g'], y=df['energy_100g'], c=df['nutriscore_grade'].astype('category').cat.codes)

Out[35]:

<matplotlib.collections.PathCollection at 0x7f64af8628e0>

In [36]:

plt.figure(figsize=(15,15))
plt.scatter(x=df['saturated-fat_100g'], y=df['energy_100g'], c=df['nutriscore_grade'].astype('category').cat.codes)

plt.figure(figsize=(15,15)) plt.scatter(x=df['saturated-fat_100g'], y=df['energy_100g'], c=df['nutriscore_grade'].astype('category').cat.codes)

Out[36]:

<matplotlib.collections.PathCollection at 0x7f64abb2a670>

In [37]:

plt.figure(figsize=(10,10))
plt.scatter(x=df['saturated-fat_100g'], y=df['nutriscore_grade'].astype('category').cat.codes)

plt.figure(figsize=(10,10)) plt.scatter(x=df['saturated-fat_100g'], y=df['nutriscore_grade'].astype('category').cat.codes)

Out[37]:

<matplotlib.collections.PathCollection at 0x7f64abb0ec70>

In [38]:

plt.figure(figsize=(10,10))
plt.scatter(x=df['fat_100g'], y=df['nutriscore_score'])

plt.figure(figsize=(10,10)) plt.scatter(x=df['fat_100g'], y=df['nutriscore_score'])

Out[38]:

<matplotlib.collections.PathCollection at 0x7f64af80c6d0>

In [39]:

plt.figure(figsize=(10,10))
plt.scatter(x=df['saturated-fat_100g'], y=df['nutriscore_score'])

plt.figure(figsize=(10,10)) plt.scatter(x=df['saturated-fat_100g'], y=df['nutriscore_score'])

Out[39]:

<matplotlib.collections.PathCollection at 0x7f64aba48880>

In [40]:

plt.figure(figsize=(15,15))
plt.scatter(x=df['fat_100g'], y=df['saturated-fat_100g'], c=df['nutriscore_grade'].astype('category').cat.codes)

plt.figure(figsize=(15,15)) plt.scatter(x=df['fat_100g'], y=df['saturated-fat_100g'], c=df['nutriscore_grade'].astype('category').cat.codes)

Out[40]:

<matplotlib.collections.PathCollection at 0x7f64ab9a9b80>

In [41]:

plt.figure(figsize=(10,10))
plt.scatter(x=df['salt_100g'], y=df['nutriscore_score'])

plt.figure(figsize=(10,10)) plt.scatter(x=df['salt_100g'], y=df['nutriscore_score'])

Out[41]:

<matplotlib.collections.PathCollection at 0x7f64ab993d90>

In [42]:

plt.figure(figsize=(10,10))
plt.scatter(x=df['sugars_100g'], y=df['nutriscore_score'])

plt.figure(figsize=(10,10)) plt.scatter(x=df['sugars_100g'], y=df['nutriscore_score'])

Out[42]:

<matplotlib.collections.PathCollection at 0x7f64ab8f06d0>

In [43]:

plt.figure(figsize=(15,15))

plt.scatter(x=df['nutriscore_score'], y=df['fat_100g'])

plt.figure(figsize=(15,15)) plt.scatter(x=df['nutriscore_score'], y=df['fat_100g'])

Out[43]:

<matplotlib.collections.PathCollection at 0x7f64ab894d60>

In [44]:

tmp_df = df.dropna()
plt.scatter(tmp_df['nutriscore_score'], tmp_df['additives_n'], alpha=0.5)
plt.xlabel('nutriscore_score')
plt.ylabel('additives_n')
plt.show()

tmp_df = df.dropna() plt.scatter(tmp_df['nutriscore_score'], tmp_df['additives_n'], alpha=0.5) plt.xlabel('nutriscore_score') plt.ylabel('additives_n') plt.show()

In [45]:

tmp_df = df.dropna()
plt.figure(figsize=(20,50))
size = tmp_df.select_dtypes('float').columns.size
for i, col in enumerate(df.select_dtypes('float')):
    plt.subplot(size, 1, i+1)
    plt.scatter(tmp_df[col], tmp_df['nutriscore_score'], alpha=0.5)
    plt.xlabel(col)
    plt.ylabel('nutriscore_score')
plt.show

tmp_df = df.dropna() plt.figure(figsize=(20,50)) size = tmp_df.select_dtypes('float').columns.size for i, col in enumerate(df.select_dtypes('float')): plt.subplot(size, 1, i+1) plt.scatter(tmp_df[col], tmp_df['nutriscore_score'], alpha=0.5) plt.xlabel(col) plt.ylabel('nutriscore_score') plt.show

Out[45]:

<function matplotlib.pyplot.show(close=None, block=None)>

Conclusion¶

This analysis on the OpenFoodFact dataset showed that it exists a correlation between the nutriscore and the nutriments. A further analysis of regression may results in some good predictions.

The dataset is huge but the data quality is not very good : lots of features are empty, lots of values are Nan or 'unknown'.